SIB text mining at TREC precision medicine 2020

Pasche, Emilie (Haute école de gestion de Genève, HES-SO // Haute Ecole Spécialisée de Suisse Occidentale ; SIB Text Mining, Swiss Institute of Bioinformatics, Geneva, Switzerland) ; Caucheteur, Déborah (Haute école de gestion de Genève, HES-SO // Haute Ecole Spécialisée de Suisse Occidentale ; SIB Text Mining, Swiss Institute of Bioinformatics, Geneva, Switzerland) ; Mottin, Luc (Haute école de gestion de Genève, HES-SO // Haute Ecole Spécialisée de Suisse Occidentale ; SIB Text Mining, Swiss Institute of Bioinformatics, Geneva, Switzerland ; University of Geneva, Switzerland) ; Mottaz, Anaïs (Haute école de gestion de Genève, HES-SO // Haute Ecole Spécialisée de Suisse Occidentale ; SIB Text Mining, Swiss Institute of Bioinformatics, Geneva, Switzerland) ; Gobeill, Julien (Haute école de gestion de Genève, HES-SO // Haute Ecole Spécialisée de Suisse Occidentale ; SIB Text Mining, Swiss Institute of Bioinformatics, Geneva, Switzerland) ; Ruch, Patrick (Haute école de gestion de Genève, HES-SO // Haute Ecole Spécialisée de Suisse Occidentale ; SIB Text Mining, Swiss Institute of Bioinformatics, Geneva, Switzerland)

TREC 2020 Precision Medicine Track aimed at developing specialized algorithms able to retrieve the best available evidence for a specific cancer treatment. A set of 40 topics representing cases (i.e. a disease, caused by a gene and treated by a drug) were provided. Two assessments were performed: an assessment of the relevance of the documents and an assessment of the ranking of documents regarding the strength of the evidence. Our system collected a set of up to 1000 documents per topic and re-ranked the documents based on several strategies: classification of documents as precision medicine-related, classification of documents as focused on the topic and attribution of a set of evidence-related scores to documents. Our baseline run achieved competitive results (rank #3 for infNDCG according to the official results): more than half of the documents retrieved in the top-10 were judged as relevant regarding the topic. All the tested strategies decreased the performances in the phase-1 assessment, while the evidence-related re-ranking improved performance in the phase-2 assessment.


Conference Type:
published full paper
Faculty:
Economie et Services
School:
HEG - Genève
Institute:
CRAG - Centre de Recherche Appliquée en Gestion
Subject(s):
Sciences de l'information
Publisher:
Virtual conference, 16-20 November 2020
Date:
2020-11
Virtual conference
16-20 November 2020
Pagination:
8 p.
Published in:
Proceedings of the Twenty-Ninth Text REtrieval Conference (TREC 2020)
External resources:
Appears in Collection:



 Record created 2021-03-15, last modified 2021-03-26

Fulltext:
Download fulltext
PDF

Rate this document:

Rate this document:
1
2
3
 
(Not yet reviewed)